(Yicai) Aug. 4 -- With domestic artificial intelligence firms such as DeepSeek having shown that an open-source approach can carve out a distinct market niche, Chinese internet giant Alibaba has announced that its flagship Qwen-Max model will also be made open-source for the first time.

Alibaba released its new-generation base model Qwen3.8-Max yesterday, offering 2.4 trillion parameters; the higher the number of parameters, the better a model can recognize patterns, generate answers, and carry out tasks.

Alibaba claimed that it is among the best large language models globally in terms of its comprehensive performance in programming, scientific research, and long-term tasks.

What makes the model stand out is Alibaba’s change of business strategy. The firm announced it plans to open-source the weights of the Qwen 3.8-Max model next week.

Since the launch of the Qwen series in 2023, the Hangzhou-based firm has released several open-source models but its Max series had previously been positioned as a closed-source flagship, accessible only via an application programming interface, without allowing users to download the model weights.

This strategic pivot comes after other Chinese firms showed that open-source, low-cost strategies can successfully capture market share. Rather than clinging to the profit potential of closed-source APIs, Alibaba is opting to pursue the open-source route to boost its influence in the future AI agent ecosystem.

Some analysts have noted that open-source strategies can hinder commercialization, but they also lower adoption barriers, expand overseas coverage, and promote cross-cloud and heterogeneous hardware deployment. Also, capable open-source alternatives will make it difficult for closed-source model providers to maintain high API pricing.

Qwen 3.8-Max costs about USD6 per million output tokens, while the cost for Anthropic’s Claude Fable 5 model is USD50. In AI agent workflows -- which require repeated tool invocation, long-form code generation, feedback loops, and continuous trial-and-error -- this disparity in costs can directly affect the commercial viability of the model.

The computing power ecosystem remains a weakness for AI startups such as DeepSeek. The overseas open-source community OpenCode said in a statement today that DeepSeek V4 Flash “is currently experiencing capacity issues due to unprecedented traffic volume; users may encounter errors, and urgent fixes are underway.”

Editor: Tom Litting